以自動音韻判讀與統計分析再探英譯《唐詩三百首》

Revisiting the Method of Analyzing Poetic Rhyme in English Translation

目前唐詩英譯研究以翻譯策略的評析與不同詩歌譯文版本的比較為主軸，鮮少關注「音韻」翻譯的層面，本研究建置了唐詩的五言與七言的絕句和律詩（共209首，絕句79首與律詩130首）之英譯語料庫，並以Python開發自動音韻判讀系統，判別英文譯本之押韻方式，包括：頭韻、諧元韻、輔音韻、視韻及尾韻之韻法，以剖析Bynner（1929）、Herdan（1973）與Xu et al.（1987）在唐詩英譯過程中的音韻轉移現象。為確保此系統開發之正確性，隨機對32首（15%）唐詩以人工進行語音標記，以進行系統評估，也就是以人工標記作為正確答案，計算精確率、召回率及調和均值三個指標。另外以統計分析進行音韻轉移與押韻方式的比較。分析結果顯示，Bynner譯本與Herdan譯本使用諧元韻及輔音韻為主，Xu et al.之譯本則大量使用了全韻，其尾韻韻法多為兩行轉韻（AABB）及隔行交互押韻（ABAB），然而三位譯者皆鮮少使用頭韻與視韻的押韻方式。本文所提出之分析方法不僅打破過往音韻翻譯只以尾韻進行分析，未來此方法可協助翻譯學者針對不同時期所出版之諸多唐詩譯本用韻方式進行共時與歷時分析。

Research on the English translation of Tang poems has focused on the evaluation of translation strategies and the comparison of different translations of poems, with little attention paid to the aural aspects of translation. In this preliminary study, a corpus of English translations of five- and seven-character quatrains and octaves of Tang poems (209 poems in total, 79 quatrains and 130 octaves) was constructed. An automated rhyme identification system was developed, and natural language processing technology was utilized to identify alliteration, assonance, consonance, eye rhyme, and perfect rhyme. This system was then employed to explore the rhyme shifting phenomenon in Bynner’s (1929), Herdan’s (1973), and Xu et al.’s (1987) English translations of Three Hundred Tang Poems. To ensure the performance of the system, we randomly selected and manually labeled 32 (15% of the corpus) Tang poems as the correct answer and applied three evaluation indices, namely precision, recall, and F1 score, to evaluate the performance of the automated rhyme identification. Further, statistical analysis was conducted to explore the differences in rhyme shifting and rhyme schema between Chinese and Western translators. According to our preliminary findings, Bynner’s and Herdan’s translations mainly use assonance and consonance, whereas Xu et al.’s translations extensively use full rhymes. The end rhymes of Xu et al.’s translations mostly comprise two-line transitions (AABB) and interlinear rhythms (ABAB). Alliteration and eye rhyme are rarely used by these three translators. The new method proposed in this paper not only breaks from previous studies that have only discussed end rhyme in rhyme-related translations but also enables translation scholars to analyze the use of rhyme in a large number of translations. Future research on the diachronic and synchronic analysis of poetic rhymes that occur in English translations is made possible by statistical methods and the automated rhyme identification system.